Does AI Favor Generalists or Specialists? What It Means for Who You Hire

By Victor Hoang · Updated Oct 10, 2026

Executives keep asking some version of one question: if AI can do a competent first draft of almost any knowledge task, who is worth hiring? The tempting answers are "generalists, because AI fills the gaps" or "specialists, because AI can't match real expertise." Both are half right, and both lead to bad hiring plans.

Our view is that AI doesn't favor either pure type. It makes competence outside a specialty cheap, which raises the value of people with one deep expertise plus real range, and lowers the premium on narrow, execution-only specialists. The rest of this piece lays out the evidence, a way to sort your roles, and what to change in your hiring this quarter.

Key Facts

• In a pre-registered experiment with 758 BCG consultants, those using AI on tasks inside AI's capability frontier completed 12.2% more tasks, finished 25.1% faster, and produced results rated more than 40% higher in quality.
• Lower performers gained the most: consultants below the average performance threshold improved by 43%, versus 17% for those above it.
• On a task deliberately placed outside the frontier, consultants using AI were 19 percentage points less likely to produce correct solutions than those without AI.
• In a field experiment with 776 professionals at Procter & Gamble, individuals with AI matched the performance of teams without AI, and R&D and Commercial staff using AI produced balanced solutions regardless of their background.
• In an experiment with 444 college-educated professionals on mid-level writing tasks, ChatGPT cut time taken by 0.8 standard deviations and raised quality by 0.4, while compressing the gap between stronger and weaker workers.

What the Studies Actually Show

It helps to be precise about who was studied, because these papers get quoted as if they were about "workers" in general.

The most-cited one is the Harvard Business School working paper on the "jagged technological frontier." Researchers worked with Boston Consulting Group and assigned 758 strategy consultants to no AI, GPT-4, or GPT-4 with a prompt-engineering overview. On 18 realistic tasks that sat inside what the model could do, AI users were faster and produced better work. The authors' point is that AI's frontier is jagged: some tasks are easy for it, while others that look equally hard are beyond it. On the one task designed to fall outside the frontier, AI users did worse than the control group, by 19 percentage points on correctness.

That second finding matters more for hiring than the first. The people who got hurt were people who trusted fluent output on a problem the tool couldn't actually solve. Knowing where the edge is takes judgment, and judgment usually comes from depth in something.

The second study, a 2025 NBER working paper called "The Cybernetic Teammate," is closer to the generalist question. It ran a field experiment with 776 professionals at Procter & Gamble working on real product-innovation challenges, alone or in pairs, with or without AI. Without AI, R&D staff proposed more technical solutions and Commercial staff proposed more commercial ones. With AI, both groups produced balanced solutions. In other words, a person with one specialty could reach acceptable work in the neighboring specialty. Individuals with AI also matched the performance of teams without it.

The third, from MIT's Shakked Noy and Whitney Zhang, studied 444 college-educated professionals doing occupation-specific writing tasks (a working paper). Time dropped by 0.8 standard deviations and quality rose by 0.4. The distribution compressed: lower-ability workers gained more, so the gap to stronger ones narrowed.

Put together, the evidence says AI raises the floor of competence, especially for people who'd otherwise be weaker at a task, and it lets people cross specialty lines with less penalty. It does not say specialists are obsolete. And none of these studies tested "spiky generalists" as a hiring category, so treat that part of our argument as an inference from the evidence, not a measured result.

The Logic Behind the Spiky Generalist

The older shorthand is the "T-shaped" professional: deep in one area, broad across others. A spiky generalist is a sharper version. One or two peaks of real expertise, a decent working range everywhere else, and, most importantly, good judgment about when the AI's output can be trusted.

Why does AI shift value toward this profile? Consider what each part of the profile now costs.

Breadth got cheaper. A marketer who needs a first-pass financial model, or an analyst who needs a contract summary, used to wait for a colleague. Now they can get to "competent draft" alone. Skills that used to require a hire or a handoff are partly commoditized at the draft level.

Depth didn't get cheaper. The BCG result shows the cost of being wrong at the frontier's edge. Someone with real expertise in an area can tell a plausible-but-wrong answer from a right one. That's the skill that keeps AI-generated work from shipping with errors.

Handoffs got more expensive by comparison. If one person can cover 70% of a cross-functional task with AI, every extra handoff to a single-function specialist adds delay and translation cost. The P&L experiment points the same direction: individuals with AI matched teams without it.

This is also why the weakest position in the market is the narrow, execution-only specialist whose value was producing a standard deliverable in one function. That's exactly the work AI drafts best. We covered the entry-level version of this problem in the broken bottom rung, and the mid-career version is similar: the premium moves from producing the artifact to judging it.

A Framework for Sorting Your Roles

Rather than ask "generalist or specialist?" in the abstract, sort each role by two questions. First: how much of the role's output is draft-level work that AI can now produce competently? Second: how costly is an undetected error in this role?

The table's last two rows deserve attention. A pure generalist can look attractive because they're flexible, and AI makes flexibility cheap, so it no longer differentiates. And a deep specialist is not threatened by AI; the specialist whose entire job was assembling routine outputs is.

Notice the pattern: the thing worth paying for is the ability to evaluate. That echoes what we've argued in the human premium and AI complementary skills.

What Changes in How You Hire

If the framework holds, a few practical shifts follow.

Screen for the peak, then probe the range. Ask a candidate to go deep on one problem they know well, then hand them an adjacent problem and watch how they use AI to get oriented. You're looking for someone who gets moving fast and can say what they'd verify.

Test judgment about AI output directly. Give candidates an AI-generated answer with a subtle error in their own area and see whether they catch it. That's the BCG outside-the-frontier failure, turned into an interview exercise. For what newer hires now expect from employers, see AI-native new hire expectations.

Stop writing job descriptions as function lists. A requirement list of eight narrow tools signals the old model. Describe the outcome, the peak skill required, and the adjacent areas where the person should be competent without a specialist's help.

Rethink the span of a team. If individuals with AI can cover more ground, smaller cross-functional pods may replace chains of single-function handoffs. The structural implications are in the org chart of AI-augmented departments.

Don't strip out the depth. The danger of the generalist narrative is hollowing out the expertise that checks the AI. Keep at least one real expert in each domain where an undetected error is expensive.

What to Do Next

For a CEO, COO, or CHRO, here is a sequence that fits one planning cycle.

1. Audit roles with the two-question table. For each open or at-risk role, score draft-level exposure and error cost. Do this for the 15 to 20 roles that make up most of your headcount cost, not all of them.
2. Freeze or reshape execution-only requisitions. Before approving a hire for a narrow, standard-deliverable role, ask whether an existing person with a deep peak and the right tools can absorb it. The upskill-versus-hire case gives the cost math.
3. Rewrite two or three job descriptions and interview loops around peak plus range plus judgment. Pilot them in one function before rolling out.
4. Name your protected specialists. List the roles where deep expertise is the control on AI errors, and make sure compensation and retention plans reflect that.
5. Build a skills view, not just a titles view. A skills-based talent strategy lets you see who has a peak and where the range is, which job titles hide.

One caveat to hold throughout: the experiments are early, they involved specific populations (consultants, product-development professionals, college-educated writers), and models keep changing. Plan for the direction, not the exact magnitudes.

Learn More

• The Human Premium
• AI Complementary Skills
• The Broken Bottom Rung
• AI-Native New Hire Expectations
• Upskill vs. Hire: The AI-Native ROI Case
• Is Experience Still Worth Paying For?
• Skills-Based Talent Strategy